By David Speights, Ph.D., and Chris Hanks, Ph.D., The Retail Equation
The economic climate is still uncertain for retailers. Although sales are improving, the National Retail Federation reports that fraudulent and abusive returns are on the rise, costing retail companies millions in profits. Additionally, shrink and organized retail crime continue to be multi-billion dollar retail problems.
As hazardous as this climate appears, it also presents an ideal opportunity for loss prevention professionals. By monitoring transactions over time and bringing statistics to bear, loss prevention analytics is reshaping operations and policies to protect bottom lines. This process often begins with “data mining”―a catch-all term for the methods analysts use to make sense of vast quantities of information. By sifting through millions of data points, analytics professionals are able to tease out relationships that would otherwise be undetectable. The result is that today’s retailers have a number of loss prevention tools that were unavailable only a few years ago. Below is an outline of some of the techniques used to maximize a retailer’s margin.
Challenging Basic Assumptions
Most retailers’ current accounting programs do not accurately reflect their real return rates; they often overlook exchange transactions and therefore understate the value and quantity of merchandise returning to the store. The return rates for 10 different retailers were recently tallied to analyze how they viewed the impact of merchandise returns. All were underestimating their return rate―one by as much as 150 percent. In fact, the 10 retailers studied saw an average return rate discrepancy of more than 80 percent.
Why is this important? Items and dollars that get returned within exchange transactions are unexpectedly hidden, masking retailers’ opportunities to rescue sales, prevent fraud, reduce shrink and more.
As a retailer, imagine that each customer who returns a product hands you a slip of paper. On the paper is written a number between 0 and 100 percent and a note that says, “This number represents the probability that my return is fraudulent.” Although not this simple, this is the end result of predictive modeling. By tracking and analyzing customers’ purchases, exchanges, and return behaviors over time, loss prevention statisticians are able to develop real-time mathematical models that accurately estimate the chances of a return being legitimate or fraudulent. Recognizing high-risk customers is important, as it often leads to broader networks of return fraud.
Beyond Exception Reporting
Bringing computing power and statistics to the process of exception reporting is a key means of reducing fraud. Today, almost all retailers’ loss prevention departments use some form of exception reporting to identify suspicious transactions, individuals or employees. This process usually involves a complex set of rules to flag certain situations that “seem” problematic.
Taking this scenario one step further, the complex rules for flagging transactions can be reduced to a set of risk variables, each of which can be correlated to known outcomes. By determining the relationship between risk variables and known outcomes (such as correlating a certain employee’s behavior with his/her ultimate termination for fraud), retailers can learn which risk attributes are most important and what weight to assign each. Feeding these variables into predictive models then yields composite risk scores for evaluating transactions, employees, stores or other units of interest. This transition from complex rules to predictive models for identifying fraudulent transactions is analogous to a transition that occurred in the 1990s in the credit card industry: improved ROI and greater loss prevention efficacy let that market do more with less.
Fraud Ring Analysis
Social analysts find that people tend to group together based on similarities, and that this is particularly true among criminals. A key method of identifying (and ultimately cracking) organized retail crime rings is by first identifying high-risk customers, and then mapping out clusters of similar customers and analyzing their transaction behavior. Using sophisticated linking algorithms such as “fuzzy matching,” loss prevention analysts can connect known fraudsters to other questionable customers, often uncovering clusters of identities that constitute either crime networks or aliases of the same criminal.
Knowing how products are associated with one another allows them to be clustered into groups and ranked for risk. Combining this information with the typical shrink data goes far beyond the groupings one might find in a standard product hierarchy. For example, consider the capability to us a common product-pairing, like a digital camera and photo paper, to create an indirectly associated product-pair, like a digital camera and a photo album. Knowing this association and crossing the information with shrink data engenders risk profiling for many products and product clusters.
Before implementing any loss prevention strategy or solution, retailers should understand both the costs and associated benefits. Controlled tests, followed by statistical analyses, aid this understanding. Using “experimental” and “control” groups of stores―and tracking key metrics such as shrink, sales, return rates, or other important outcomes in before-during-after analyses―loss prevention professionals can accurately calculate a given strategy’s ROI. Controlled trials also let analysts manipulate elements that make up an overall strategy: By correlating changes in strategy with changes in ROI, statisticians can optimize loss prevention policies.
Clearly, statistics play a growing role in retailers’ approach to loss prevention issues and solutions. This is important in any economic climate; but in a mixed economy where profit margins are uncertain, it is imperative for retailers to have an unambiguous picture of their business that is rooted in solid statistical analysis.
David Speights, Ph.D., is the chief statistician and Christopher Hanks, Ph.D., is the senior statistician of The Retail Equation, the industry leader in retail transaction optimization solutions. The company’s applications use statistical modeling and analytics to predict consumer behavior and turn each individual shopper visit into a more profitable experience. Its software-as-a-service applications operate in more than 15,000 stores in North America, supporting a diverse retail base of specialty, department, sporting goods, auto parts and more. For more information, visit www.theretailequation.com.